This invention relates to a laser beam homogenizer and, more particularly, to a single element binary diffractive optical element beam homogenizer.
Lasers have a wide range of applications involving etching and shaping of a variety of materials by annealing or ablation. For example, lasers can be used in semiconductor device fabrication, microphotolithography and photodeposition. Lasers provide an intense light source which can be precision controlled to etch and shape materials.
However, the effectiveness of a laser in annealing and ablation requires a uniform laser beam intensity profile to produce a smooth and even energy distribution across the material. Ideally, the intensity profile should be rectangular with straight vertical sides and a flat horizontal top.
When a laser is used, there often are variations in the etching and shaping across the resulting material. The variations stem from the non-uniform intensity distribution of the laser light. Generally, the intensity of a laser beam has a spatially non-uniform distribution, peaking at the center of the beam and gradually decreasing towards the circumference of the beam.
A homogenizer is often used to achieve better uniformity in light intensity. The laser beam in a typical homogenizer is often divided into multiple sections, which are then redirected to different positions and overlapped back together to produce a beam with a more uniform intensity profile distribution. This process is referred to as "beam-folding".
However, the overlapping beams, resulting from beam-folding, interfere with each other to form a standing wave pattern which creates an intensity modulation or an interference fringe pattern. This intensity modulation in turn creates non-uniformities in the interference pattern on the material.
This intensity modulation problem, also referred to as the "standing wave problem", has been widely observed in photolithography processes using coherent light sources such as lasers. The intensity modulation causes non-uniformity from one location to another in the degree to which a photoresist layer is exposed. One approach to resolve this problem in the photolithography area is to use a homogenizer in which the laser beam moves or scans the target material during the exposure. Since multiple pulses are typically required to expose a photoresist area, each pulse in the series of pulses is spatially offset relative to the next pulse. Better uniformity when exposing photoresist is achieved by moving the laser beam during an exposure because the interference fringe patterns of all the pulses are averaged out.
An example of such an approach to resolve the standing-wave problem is described in U.S. Pat. No. 5,153,773 entitled "Illumination Device Including Amplitude-Division and Beam Movement" invented by Muraki, which teaches the use of a scanning fly-eye homogenizer. The amplitude of the light from a laser source is first divided into two light beams by a polarization beam-splitter, after which the light beams are inputted to a rotating wedge prism which is used to shift their paths and their positions of incidence. By imparting an optical path difference and by shifting the positions of the beams, mutually incoherent light beams are produced to avoid or to minimize the formation of an interference fringe pattern at the material.
However, typical techniques used to improve laser beam intensity uniformity in photolithography processes are generally not applicable to the crystallization of semiconductor materials. If a photoresist layer is irradiated by multiple pulses, the integrated cumulative irradiation of all the pulses would determine the degree to which the photoresist has been exposed. During the exposure process, each laser pulse incrementally increases the degree to which the photoresist has been exposed.
If multiple pulses are applied to a semiconductor target such as silicon, the total energy transferred to the silicon does not determine the material and electrical characteristics of the silicon target. Because silicon has a fast and non-linear response to light intensity or process fluence, the characteristics of the target are primarily determined by the energy output of the last pulse upon the sample. The response time of silicon is so fast that during each pulse, the silicon film will melt and recrystallize. Since the pulse length of a laser shot is approximately 30 nanoseconds, homogenizers which use a moving optical part to smear out the interference patterns of the pulses such as described in the Muraki patent are not generally applicable to crystallization of semiconductor materials. Laser crystallization of silicon requires the intensity of each pulse to be homogeneous because of the fast response time of silicon. However, mechanical parts generally cannot rotate or move sufficiently fast to achieve single-pulse homogeneity.
Current beam homogenizers used for prior art laser applications are plagued with several problems. In their simplest form, a homogenizer 10 of FIG. 1 consists of a beam expander lens 12, a beam condenser 14 and a biprism pair 16. A laser light source 18 will emit a coherent light beam 20 which is expanded and contracted by lenses 12 and 14 to achieve a beam of the correct aspect ratio for the biprism pair and divided into four slightly converging beams 22, 24, 26, and 28 by the biprism pair 16, which are then folded into a single beam 30 at the image plane 32.
The four slightly converging coherent light beams form a standing wave pattern which produces an interference fringe pattern in the beam 30 at the image plane 32. This interference fringe pattern produces an intensity distribution, I, which varies in the x and y direction according to EQU I(x,y).about.[1+cos(2 kx sin(.theta..sub.x))][1+cos(2 ky sin(.theta..sub.y))] Equation1
where k is the wavenumber, and .theta. describes the tilt angle (or convergence angle) induced by the biprism 16 in the x and y directions. As an example, a commercially available homogenizer induces a beam tilt of 13 mrad in the x and y directions, resulting in an interference pattern with a 12 .mu.m pitch. This 12 .mu.m pitch is larger than a thermal diffusion length during the solidification of silicon during pulsed laser crystallization of amorphous silicon. The result is a highly inhomogeneous distribution of grain sizes caused by the rapid spatial variation in fluence. The biprism homogenizer homogenizes macroscopically and dehomogenizes mesoscopically.
The second problem with conventional homogenizers is that the abrupt splitting of the laser beam by the prism faces of the biprism causes edge Fresnel diffraction. The result is predicted by theory to be a 37% spiking of the beam at the edges as shown in FIG. 2 which is in close agreement with actual beam profiles of FIG. 3. FIG. 2 shows the intensity profile of a collimated optical wavefront which has undergone edge diffraction. Rather than being uniformly dark in the shadow region, and uniformly intense in the illuminated clear aperture region, the wavefront diffracts to produce Fresnel oscillations. These oscillations are visible at the edges of a measured beam profile taken on a beam homogenized with a biprism homogenizer. Instead of an edge which abruptly terminates a wavefront, the biprism apex abruptly interrupts the phase profile of the wavefront. The resulting Fresnel edge diffraction is essentially the same in either case. As a practical matter, the Fresnel diffraction undermines the macroscopic homogeneity of the beam profile.
The third problem with conventional homogenizers is that the "beam-folding" operation is not flexibly tailored to the original beam profile. The biprism pair can at best overlap four pieces of the beam profile. This is too inflexible to macroscopically homogenize all but the simplest beam profile functions.
The propagation of a light beam can be changed by three basic means: reflection by a mirror, refraction by a lens and diffraction by a grating. Homogenizers traditionally rely on reflection and refraction as noted to achieve the desired optical transformation. Optical design, based on mirror and lens elements, is a well-established and refined process. Until recently, the problems with diffraction and fabricating high efficiency diffractive elements have made diffractive elements unfeasible components of optical systems.
The diffractive process does not simply redirect a light beam. Diffraction, unlike refraction and reflection, splits a light beam into many beams--each of which is redirected at a different angle or order. The percentage of the incident light redirected by the desired angle is referred to as the diffraction efficiency. The diffraction efficiency of a diffractive element is determined by the element's surface profile. If the light that is not redirected by the desired angle is substantial, the result will be an intolerable amount of scatter in the image or output plane of the optical system.
A diffractive optical echelon has been proposed as a beam-smoothing means for a laser beam in U.S. Pat. No. 4,521,075 to Obenschain et al. Ten or more transparent plates of random heights are stacked together to break a coherent beam into multiple different delayed beamlets. The random beamlets are then focused by a separate condenser lens and overlapped so that the interference pattern of the resulting beam will average out to a smooth diffraction profile.
However, the Obenschain et al. patent requires ten or more transparent stacked plates. It requires random heights to those plates to delay the transmitted beamlets and it trusts to the law of averages to provide a uniform profile beam.
Theoretically, on-axis diffractive phase elements can achieve 100 percent diffraction efficiency. To achieve this efficiency, however, a continuous phase profile is necessary. The theoretical diffraction efficiency of this surface profile is also relatively sensitive to a change in wavelength. By contrast, refractive elements are wavelength insensitive. The technology for producing high quality, high efficiency, continuous phase profiles does not presently exist.
A compromise that results in a relatively high diffraction efficiency and ease of fabrication is a multi-level phase grating binary diffractive optical element. The larger the number of discrete phase levels, the better the approximation of the continuous phase function. These multi-level phase profiles can be fabricated using standard semiconductor integrated circuit fabrication techniques.
As disclosed in Binary Optics Technology: The Theory and Design of Multi-level Diffractive Optical Elements by G. J. Swanson of the Lincoln Laboratory at the Massachusetts Institute of Technology, (Technical Report 854, Aug. 14, 1989) and the resulting U.S. Pat. No. 4,895,790, a fabrication process starts with a mathematical phase description of a diffractive phase profile and results in a fabricated multi-level diffractive surface. The first step is to take the mathematical phase expression and generate from it a set of masks that contain the phase profile information. The second step is to transfer the phase profile information from the masks into the surface of the element specified by the lens design.
The first step involved in fabricating the multi-level element is to mathematically describe the ideal diffractive phase profile that is to be approximated in a multi-level fashion. The next step in the fabrication process is to create a set of lithographic masks which are produced by standard pattern generators used in the integrated circuit industry.
A substrate of the desired material, such as Ge, ZnSe, Si, and SiO.sub.2, is coated with a thin layer of photoresist. A first lithographic mask is then placed in intimate contact with the substrate and illuminated from above with an ultraviolet exposure lamp. Alternately, pattern generators, either optical or electron beam, can expose the thin layer of photoresist. The photoresist is developed, washing away the exposed resist and leaving the binary grating pattern in the remaining photoresist. This photoresist will act as an etch stop.
The most reliable and accurate way to etch many optical materials is to use reactive ion etching. The process of reactive ion etching anisotropically etches material at very repeatable rates. The desired etch depth can be obtained very accurately. The anisotropic nature of the process assures a vertical etch, resulting in a true binary surface relief profile. Once the substrate has been reactively ion etched to the desired depth, the remaining photoresist is stripped away, leaving a binary phase surface relief grating.
The process may be repeated using a second lithographic mask having half the period of the first mask. The binary phase element is recoated with photoresist and exposed using the second lithographic mask which has half the period of the first mask. After developing and washing away the exposed photoresist, the substrate is reactively ion etched to a depth half that of the first etch. Removal of the remaining photoresist results in a 4 level approximation to the desired profile. The process may be repeated a third and fourth time with lithographic masks having periods of one-quarter and one-eighth that of the first mask, and etching the substrates to depths of one-quarter and one-eighth that of the first etch. The successive etches result in elements having 8 and 16 phase levels. More masks than four might be used, however, fabrication errors tend to predominate as more masks are used.
This process is repeated to produce a multilevel phase relief structure in the substrate. The result is a discrete, computer-generated structure approximating the original idealized diffractive surface. For each additional mask used in the fabrication process, the number of discrete phase levels is doubled, hence the name "binary" optical element or, more precisely, a binary diffractive optical element.
After only four processing iterations, a 16 phase level approximation to the continuous case can be obtained. The process can be carried out in parallel, producing many elements simultaneously, in a cost-effective manner.
A 16 phase level structure achieves 99 percent theoretical diffraction efficiency. The residual 1 percent of the light is diffracted into higher orders and manifests itself as scatter. In many optical systems, this is a tolerable amount of scatter. The fabrication of the 16 phase level structure is relatively efficient due to the fact that only four processing iterations are required to produce the element.
After the first etching step, the second and subsequent lithographic masks have to be accurately aligned to the existing pattern on the substrate. Alignment is accomplished using another tool standard to the integrated circuit industry, a mask aligner.
As noted, the photoresist on the substrate can be exposed with an electron-beam pattern generator. The e-beam direct-write process eliminates masks and their corresponding alignment and exposure problems. Binary optics have also been reproduced using epoxy casting, solgel casting, embossing, injection molding and holographic reproduction.
Binary optical elements have a number of advantages over conventional optics. Because they are computer-generated, these elements can perform more generalized wavefront shaping than conventional lenses or mirrors. Elements need only be mathematically defined: no reference surface is necessary. Therefore, wildly asymmetric binary optics are able to correct aberrations in complex optical systems, and elements can be made wavelength-sensitive for special laser systems.
The diffractive optical elements are generally thinner, lighter and can correct for many types of aberrations and distortions. It is possible to approximate a continuous phase profile with a stepwise profile of discrete phase levels.
It is an object of this invention to provide a single inexpensive optical element as a beam homogenizer.
It is a further object of this invention to provide a binary diffractive optical element as a beam homogenizer.